Skip to content
Conduit

skill

Prdfy

Pre-coding agent skill: an AI orchestrator interviews you about your product idea and produces a stack-agnostic specification — business plan, PRD, roadmap, risk matrix, and C4 architecture — before a

abbasmhd/Prdfycommunitylistedunreviewed

Set up

Skills are installed by placing their folder where your agent loads skills.

The download comes from github.com/abbasmhd/Prdfy itself, not from our server, so what you get is exactly what its author published.

Install instructions come from the README

Skills do not use an MCP config. Follow the steps in the README below.

README

synced Oct 22, 2026

Prdfy

License: MIT

Prdfy (Predify) is a pre-coding skill. It interviews you about a software product idea and writes the specification before any implementation starts.

You talk to one orchestrator. Sub-agents do the research, the interview, the writing, and the review. When a sub-agent needs a fact only you know, it stops. The orchestrator asks you, then sends your answer back to that sub-agent.

The specification stays stack-agnostic. It names roles such as Interactive User Client, Relational State Store, and Message Broker. It does not choose a language, framework, or database.

Interview

The orchestrator runs six modules, in order, two to four questions at a time:

  1. Vision
  2. Scope
  3. Domain Rules
  4. Strategy and MVP
  5. Risk and Compliance
  6. Architectural Boundaries

During the interview, a research sub-agent looks up competitors, market dynamics, and regulations. Claims are labeled as coming from you, from research, or from an explicit assumption.

Deliverables

After the architectural boundaries are confirmed, writers produce eight documents and an index in prdfy-spec/:

FileDeliverable
00-decision-log.mdDecisions and rules captured during the interview
01-business-plan.mdBusiness plan and value proposition
02-prd.mdProduct requirements document
03-specifications.mdSpecifications and domain contracts
04-strategy-roadmap.mdStrategy and product roadmap
05-risk-compliance.mdRisk management and compliance matrix
06-domain-architecture.mdDomain architecture
07-c4-architecture.mdC4 model, levels 1–4, in Mermaid
README.mdIndex of the documents, open assumptions, and sources

A verifier checks each file, then checks that the documents agree with each other. The skill stops at the specification.

The run state lives in prdfy-spec/memory.md. That file records your rules, decisions, answers, assumptions, and progress, so you can stop and pick up the run in a later session. Research, module, and verification notes go to prdfy-spec/findings/.

Install

The skill is seven folders under .agents/skills/: the prdfy orchestrator and its six sub-agent skills (prdfy-research, prdfy-module, prdfy-writer, prdfy-c4, prdfy-verifier, prdfy-review). Install all seven, from the same version. The orchestrator checks that every skill declares the same metadata.protocol, and it stops if one differs. The host needs sub-agents, web search, and Mermaid. After installing, start a new Agent chat and invoke /prdfy, or ask the agent to scope a product idea.

npx skills

List what the repo offers:

npx skills add abbasmhd/Prdfy --list

Install all skills for Cursor in the current project:

npx skills add abbasmhd/Prdfy --skill '*' -a cursor -y

Install all skills for Cursor on every project on this machine:

npx skills add abbasmhd/Prdfy --skill '*' -a cursor -g -y

Install all skills for every agent on this machine:

npx skills add abbasmhd/Prdfy --all

For Cursor, a project install lands in .agents/skills/. A global install (-g) lands in ~/.cursor/skills/. --all is global across agents and does not take -a. Leave off -y if you want the CLI to ask which skill, which agent, and whether to symlink or copy.

To install from a local checkout instead:

npx skills add /path/to/Prdfy --skill '*' -a cursor -y

If a symlink fails on Windows, add --copy. Other agents use the same command with their own --agent name, such as claude-code.

This repository

The skill is already installed for anyone who opens this repo in Cursor. Ask the agent to brainstorm, scope, or architect a product idea, or invoke /prdfy.

Another project

From this repo, copy the folder into the other project:

mkdir -p /path/to/other-project/.agents/skills
cp -R .agents/skills/prdfy .agents/skills/prdfy-research .agents/skills/prdfy-module .agents/skills/prdfy-writer .agents/skills/prdfy-c4 .agents/skills/prdfy-verifier .agents/skills/prdfy-review /path/to/other-project/.agents/skills/

Cursor also loads skills from .cursor/skills/. Claude Code loads them from .claude/skills/. Use whichever directory that project already uses, and copy all seven folders into it.

Every project on your machine

Copy it into your user skills directory:

mkdir -p ~/.agents/skills
cp -R .agents/skills/prdfy .agents/skills/prdfy-research .agents/skills/prdfy-module .agents/skills/prdfy-writer .agents/skills/prdfy-c4 .agents/skills/prdfy-verifier .agents/skills/prdfy-review ~/.agents/skills/

The same seven folders work under ~/.cursor/skills/ for Cursor and ~/.claude/skills/ for Claude Code. On Windows, ~ is your user profile (C:\Users\<you>).

User-level skills stay on that machine. Commit the project copy if teammates should get the skill with the repo.

Other agents

Point the agent at SKILL.md, or paste its body into the agent instructions. Keep one orchestrator in front of the user and delegate every step to a sub-agent.

Start from a product idea, not from a request to write code. If a technology is already chosen, the skill records it as your constraint and still describes the architecture in role names. If the host cannot launch sub-agents, the orchestrator says so and runs each step inline as a separate role.

Development

The skills need no runtime. The repository has two check scripts. Run them with uv:

# Role lists and protocol versions agree across the seven skills
uv run scripts/check_skills.py

# Static checks on a generated specification folder
uv run scripts/check_spec.py path/to/prdfy-spec

# Tests for both scripts, against tests/fixtures/minimal-spec
uv run --with pytest --with pyyaml pytest tests

CI runs these checks on every push and pull request to main.

Evals for the orchestrator live in .agents/skills/prdfy/evals/:

  • evals.json holds the end-to-end scenarios, in the skill-creator format.
  • trigger-evals.json holds the description trigger set, for skill-creator's run_eval.py and run_loop.py. run_eval.py and improve_description.py launch the command in PRDFY_AGENT_CMD (or --agent-cmd), and {query} marks where the query or proposal prompt goes. If neither is set, each script exits with an error that names PRDFY_AGENT_CMD and does not launch a built-in client or invent a description.

License

Prdfy is licensed under the MIT License.

↑↓ navigate openesc close